Anthropic, OpenAI must now watermark AI-generated content in Europe or risk $17m penalties

Europe is now fining AI companies that don’t watermark their content

The era of voluntary AI watermarking promises is over — Europe has made it the law. Under Article 50 of the EU AI Act, technology companies are now required to embed machine-readable watermarks in AI-generated content, with non-compliance carrying fines of up to €15 million ($17 million) or 3% of global annual turnover, according to Politico Europe.

Anthropic has already begun watermarking content generated by Claude models launched in the EU on or after August 2, 2026, with existing models to follow. Watermarking is applied at the model level, meaning it is present regardless of which Claude product or surface the text comes from.

OpenAI, meanwhile, is adopting a layered approach combining standards, watermarking and verification tools, acknowledging that no single detection method is perfect and that watermarks can sometimes be removed or fail to survive edits or platform changes.

The legislative shift follows years of voluntary pledges from major tech companies dating back to 2023 and 2024, driven by deepfake-fuelled misinformation campaigns that threatened election integrity across multiple countries. The EU’s move is now forcing the industry’s hand globally.